1. The achievable throughput of the various architectures from implementing
The DFT is to be estimated. To simplify their evaluation, it is assumed that the
delay characteristics of the multipliers and adders dominate and that the contributions
to the delay from other components such as registers and multiplexers
for example, are negligible in comparison. The delay of an n-bit ad-
tier is given to be n - T0 and that of an n X n-bit multiplier 3n – T0. The input
values and the coefficients are each given in component form with m bits bits.
The adder for accumulation is constructed such that the intermediate results
can be repesented with a maximum of 3mbit bits. The necessary reduction
to 3 m bit bit occurs via truncation.
a. the critical delays of a PE as in Figure 7.2.2 and of a butterfly PE as in
Figure 7.3.2 are to he calculated for the general case.
b. Which throughput can be achieved through use of one PE? The following
particular values are to be assumed:
mbit = 8, N= 64, T9 = 0.5 ns.
c. How much does the throughput increase through use of l-D arrays in
accordance with Figures 7.2.1, 7.3.5 and 7.3.10?